ConvIDS: A Convolutional LSTM Based Intrusion Detection Model for In-Vehicle CAN Bus

With the popularity of intelligent vehicles, there are numerous concerns regarding the security of the in-vehicle networks (IVNs). Controller Area Network (CAN) is one of the primary networks utilized in vehicles due to its high reliability and low cost. However, the absence of security mechanisms makes it vulnerable to cyber attacks. Once the attacker successfully gains access to the target CAN bus through the physical or wireless interface, various types of injection attacks can be conducted. To detect malicious attack on communication of the CAN bus, an efficient intrusion detection system (IDS) is required. Since the data of the CAN bus is typically in the form of time series, recurrent neural network (RNN) based methods have been proven effective for in-vehicle intrusion detection. In this paper, we proposed a novel approach named ConvIDS. The method is based on CAN image transforming and Convolutional Long Short-Term Memory Network (ConvLSTM). The experiments on real vehicle datasets demonstrate that the proposed model outperforms traditional neural network based methods in aspects of all metrics.